PhD/Postdoc Position for Provably Safe Reinforcement Learning

13.10.2025, Wissenschaftliches Personal

The research group Cyber-Physical Systems of Prof. Matthias Althoff at the Technical University of Munich offers a PhD/Postdoc position in the area of provably safe reinforcement learning. The Technical University of Munich is one of the top research universities in Europe fostering a strong entrepreneurial spirit and international culture.

Expected Starting Date: December 2025 - April 2026

Closing Date for Applicants: 15 November 2025

Duration: 3 years with a possible extension (individual duration for Postdocs).

Project and Job Description

Data-driven approaches are gaining importance for autonomous vehicles. However, the training and certification of autonomous systems with machine learning components is a huge challenge, since the learned behavior is often represented in large neural networks that are hard to analyze and whose decision processes cannot be interpreted by humans. To make this technology available without sacrificing safety concerns, we propose to combine machine learning techniques with formal methods.

We will focus on safe reinforcement learning of motion planning problems for autonomous vessels. Motion planning is particularly interesting, because the motion does not only have to be safe (i.e., collision-free), but also has to comply with rules. The focus on reinforcement learning is motivated by the fact that it does not require labeled data and that the learned intelligent agents can perform better than humans. Instead of verifying the correctness of neural networks, we build a safety net that only forwards safe actions to the actuators of the vehicle. This has the critical advantage that one can easily replace machine learning components within the safety net, without having to re-certify the system. In addition, the safety net only verifies if the currently proposed actions are safe given the current state of the environment in an online fashion. Although this procedure has to be performed online, it is a much easier verification problem compared to verifying a neural network with millions of neurons. We will also use the feedback of the safety layer to improve the learning rate by excluding provably unsafe actions during exploration.

These methods will be developed together with the company CargoKite (https://cargokite.com/), which develops a ship for autonomous, highly flexible global container transportation. The transport ship uses kite-based propulsion and an additional diesel engine to bridge port trips and wind lulls. Appropriate interfaces are also being developed for the new control system so that, in addition to its use in the CargoKite container ship, a stand-alone package is created that can also be used for other wind-powered and purely motor-powered ships and will be marketable as an independent product. The new system to be developed will be tested and optimized with an existing CargoKite ship prototype in real operation.

Previous Work

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10530091

https://mediatum.ub.tum.de/doc/1687691/ltbbjgwihiht96fqfeblxdyj7.ITSC22_CommonOcean.pdf

Job Specifications

• For PhD applicants: Excellent Master’s degree (or equivalent) in computer science, engineering, or related disciplines (typically mathematics, physics).

• For Postdoc applicants: Excellent track record in computer science or engineering.

• Fluency in spoken and written English is required.

• Proficient in at least one programming language, e.g. MATLAB, C/C++, Python.

• Highly motivated and keen on working in an international and interdisciplinary team.

• Applicants with strong background in the following fields are preferred:

  • Machine Learning
  • Formal Methods
  • Dynamical Systems
  • Control Theory

Context

The applicant will be directly advised by Prof. Matthias Althoff (https://www.ce.cit.tum.de/cps/members/prof-dr-ing-matthias-althoff/). Besides excellent skills for conducting innovative science, the candidate should also be talented in implementing research results on a real ship and lead teams of students.

Our Offer

PhD remuneration will be in line with the current German collective pay agreement TV-L E13 (around 4600 Euros/month in the first year, 4900 Euros/month second year). The Technical University of Munich is an equal opportunity employer committed to excellence through diversity. We explicitly encourage women to apply and preference will be given to disabled applicants with equivalent qualifications.

Contact

International candidates are highly encouraged to apply. Please submit your complete application (in English or German) via our application form: https://wiki.tum.de/display/cpsforms/Ph.D.+Application. Fill out all mandatory fields (*) and kindly use “Provably Safe Reinforcement Learning” as the “Title of Position”. Please do not include a cover letter.

Further similar job offerings will be announced on https://www.ce.cit.tum.de/cps/open-positions/.

Kontakt: tobias.ladner@tum.de